| Summary: | Thermokarst lakes are widely distributed on the Qinghai-Tibet Plateau (QTP). However, owing to the lack of high-precision remote sensing imagery and the difficulty of in situ monitoring of permafrost regions, quantifying the changes in the distribution of thermokarst lakes is challenging. In this study, we used four machine learning methods—random forest (RF), gradient boosting decision tree (GBDT), classification and regression tree (CART), and support vector machine (SVM)—and combined various environmental factors to assess the distribution of thermokarst lakes from 2015 to 2020 via the Google Earth Engine (GEE). The results indicated that the RF model performed optimally in the extraction of thermokarst lakes, followed by GBDT, CART, and SVM. From 2015 to 2020, the number of thermokarst lakes increased by 52%, and the area expanded by 1.6 times. A large proportion of STK lakes (with areas less than or equal to 1000 m2) gradually developed into MTK lakes (with areas between 1000 and 10,000 m2) in the central part of the QTP. Additionally, thermokarst lakes are located primarily at elevations between 4000 and 5000 m, with slopes ranging from 0 to 5°, and the sand content is approximately 65%. The normalized difference water index (NDWI) and enhanced vegetation index (EVI) were the most favourable factors for thermokarst lake extraction. The results provide a scientific reference for the assessment and prediction of dynamic changes in thermokarst lakes on the QTP in the future, which will have important scientific significance for the studies of carbon and water processes in alpine ecosystems.
|
|---|